NeurIPS 2020

Model Agnostic Multilevel Explanations

Meta Review

Overall, the reviewers like the contributions of the paper and feel it will be a valuable contribution. However, they strongly urge that the authors describe the limitations of the work, in particular, the scalability issues that arise from the need to tune the alpha-i for each instance, and clarify the concern with using pre-defined graphs (and how it can significantly bias the results). These revisions will make the paper substantially stronger.